Affordable Early Detection of Autism Spectrum Disorder Using Innovative Eye-Tracking Methods

We propose a cost-effective and accessible approach for early detection of Autism Spectrum Disorder (ASD) using eye-tracking data and machine learning models. Traditional methods like functional magnetic resonance imaging (fMRI) and behavioral surveys are often costly, time-consuming, and difficult to implement widely. Existing eye-tracking techniques are typically confined to controlled lab environments. To address these challenges, we utilize a comprehensive eye-tracking dataset that includes eye fixation points, pupil sizes, and gaze coordinates, collected through portable eye-tracking goggles. The data is preprocessed (cleaned and normalized) before classification with models like Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Support Vector Machines (SVM), and k-Nearest Neighbors (KNN). Among these, CNN achieved the highest accuracy at 93.9%. By simplifying the detection process, this study offers a secure, affordable, and scalable solution for ASD screening, potentially transforming early diagnosis and intervention.